{
  "id": 4635,
  "url": "https://arxiv.org/abs/2605.09314v1",
  "title": "How LLMs Are Persuaded: A Few Attention Heads, Rerouted",
  "summary": "Language models can be persuaded to abandon factual knowledge. This vulnerability is central to AI safety, but its internal mechanism remains poorly understood. We uncover a compact causal mechanism for persuasion-induced factual errors. A small set of mid-layer attention heads almost entirely determines the model's answer. These heads write answer options into a low-dimensional polyhedron, with options occupying distinct vertices. Persuasion does not blur belief or merely reduce confidence; it ",
  "authors": "Xiangkun Sun, Lingkai Kong, Aoqi Zhang, Liang Zeng, Tonghan Wang",
  "category": "research",
  "topics": "safety-alignment",
  "orgs": null,
  "regions": null,
  "published_at": "2026-05-10T04:15:24.000Z",
  "fetched_at": "2026-07-14T16:31:08.356Z",
  "source_slug": "arxiv-ethics",
  "source_name": "arXiv",
  "source_homepage": "https://arxiv.org",
  "ethics_ai_record_url": "https://ethics.ai/record/4635",
  "original_url": "https://arxiv.org/abs/2605.09314v1",
  "evidence_status": "source-only",
  "attribution": "via ethics.ai"
}